Cities and tech companies are racing to build massive AI clusters, betting scale will drive faster model training, lower costs and wider adoption. In China, computing facilities have emerged as a new form of infrastructure over the past two years, sparking a race among cities and technology companies to build 10,000-card computing clusters. These clusters – which link 10,000 or more artificial intelligence accelerator chips – enable faster iteration of AI capabilities and significantly reduce model training times. Domestic champions, such as Huawei Technologies, Alibaba Group Holding and Moore Threads, are competing to position their chips at the center of these systems.
A 10,000-card cluster functions as a supercomputer, integrating high-performance GPUs and advanced storage into a single system. Given their scale and complexity, only a limited number of domestic providers can deliver chips for such systems. A recent note from Orient Securities identified Huawei, Hygon Information Technology and Moore Threads as first-tier players. “Having the capability to deliver clusters with tens of thousands of cards signifies a firm foothold in the most intensely competitive area of Sino-U.S. tech competition,” the report said.
Cities across China are vying to host these projects. Many are led by the country’s three state-backed telecoms operators, often in collaboration with partners across the value chain. ByteDance and Alibaba are also building private clusters to support internal AI development and provide cloud services. Alibaba, for example, recently deployed a 10,000-card cluster powered by its Zhenwu AI chips in Shaoguan, Guangdong province, in partnership with China Telecom. This followed the launch of a 10,000-card intelligent computing cluster, built with Huawei’s Ascend 910C AI chips in Shenzhen in March.
Elsewhere, Harbin has emerged as a northern hub. China Mobile launched the country’s largest single-operator intelligent computing center there in late 2024, with more than 18,000 domestic AI accelerator cards. In central China, Zhengzhou activated a 60,000-card cluster in April, built by state-backed supercomputer maker Sugon for scientific AI applications.
A key trend is the shift towards home-grown chips, part of China’s broader push for semiconductor self-sufficiency in response to U.S. export controls. The Huawei-powered cluster in Shenzhen is the first in China fully equipped with domestically developed advanced chips. Government, finance and state-owned enterprises were expected to lead uptake due to data sovereignty requirements and strong budgets, he said. The next wave was likely to come from healthcare and education. Universities were expected to rent computing power from these clusters to support research. As AI evolves towards multimodal systems, model sizes are expanding rapidly, with parameter counts growing tenfold each year. Analysts expect cluster sizes to scale beyond 100,000 – and potentially up to one million – cards, the South China Morning Post reports.
